Using Portable Consumer-Grade EEG Data as a Neurophysiological Biomarker for Cognitive Impairment Assessment in real-world scenarios through Machine Learning

The early detection of dementia represents acritical challenge in aging populations worldwide. Indeed, tra-ditional cognitive tests may lead to misdiagnosis at early stages.Biomarkers like cerebrospinal fluid (CSF) can be invasive,and neuroimaging approaches might be costly or with limitedavailability. Electroencephalography (EEG) is cost-effective,inexpensive and can become ubiquitous with the advancementof technology. Event-related evoked potentials (ERP – P300)can be valuable for dementia assessment. However, despite therecent advancement, this technology has not translated intoclinical tools, and even less into commonly used technology bylaymen in daily-life activities.We designed a framework the manage the datastreaming ina federated learning manner, for both an EEG-based passiveP300 application, and a big data scenario following multiplepeople with consumer-grade EEG device to identify EEG-basedfeatures and stratify subjects according. The P300 applicatio isdesigned to assess how emotional stimuli of different valencescan impact the modulation of visual attention. Its primaryobjective is to assess cognitive impairment using classicalneurophysiological metrics (P300 latency and amplitude) andapply machine learning models for differentiation.We conducted a cross-sectional study with elderly partici-pants recruited from senior houses in Kraków, Poland. Partici-pants underwent standardized cognitive assessments. EEG datawere recorded using the MUSE portable device and analyzedusing advanced machine learning algorithms. This researchaims to validate the feasibility of portable EEG devices asaccessible screening tools for cognitive decline in communitysettings. A machine learning based subsystem was used toclassify how EEG responses vary across these different groupsaiding the differential diagnosis.